Abstract
Detection of plant disease is usually difficult without experts' knowledge. Therefore, fast and accurate automated method has been highly desired in agricultural fields.Several studies on automated plant diagnosis have been investigated using machine learning methods, such as self-organization maps or neural networks. However, these methods have been faced with several difficulties, i.e. detection of ROI, design and implementation of efficient parameters and so on.In this study, we present a novel detection system for plant disease using convolutional neural networks (CNN). CNN automatically acquires requisite features for the classification only from the training images and achieves high classification performance. We used a total of 800 cucumber leaf images and trains CNN with the distinctive investigations on its training and estimation phases.The proposed system with our well thought out CNN achieved the classification accuracy of 88.1%, sensitivity of 84.4% and 90.5% for each diseases under the 4-fold cross-validation strategy.
